summaryrefslogtreecommitdiff
path: root/lib/Target/AMDGPU/MCTargetDesc/SIMCCodeEmitter.cpp
blob: bda0928036fdeccea63121580aecf0cb19acd007 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
//===-- SIMCCodeEmitter.cpp - SI Code Emitter -----------------------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
/// \file
/// \brief The SI code emitter produces machine code that can be executed
/// directly on the GPU device.
//
//===----------------------------------------------------------------------===//

#include "AMDGPU.h"
#include "MCTargetDesc/AMDGPUFixupKinds.h"
#include "MCTargetDesc/AMDGPUMCCodeEmitter.h"
#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
#include "Utils/AMDGPUBaseInfo.h"
#include "llvm/MC/MCCodeEmitter.h"
#include "llvm/MC/MCContext.h"
#include "llvm/MC/MCExpr.h"
#include "llvm/MC/MCFixup.h"
#include "llvm/MC/MCInst.h"
#include "llvm/MC/MCInstrDesc.h"
#include "llvm/MC/MCInstrInfo.h"
#include "llvm/MC/MCRegisterInfo.h"
#include "llvm/MC/MCSubtargetInfo.h"
#include "llvm/MC/MCSymbol.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include <cassert>
#include <cstdint>
#include <cstdlib>

using namespace llvm;

namespace {

class SIMCCodeEmitter : public  AMDGPUMCCodeEmitter {
  const MCRegisterInfo &MRI;

  /// \brief Encode an fp or int literal
  uint32_t getLitEncoding(const MCOperand &MO, const MCOperandInfo &OpInfo,
                          const MCSubtargetInfo &STI) const;

public:
  SIMCCodeEmitter(const MCInstrInfo &mcii, const MCRegisterInfo &mri,
                  MCContext &ctx)
      : AMDGPUMCCodeEmitter(mcii), MRI(mri) {}
  SIMCCodeEmitter(const SIMCCodeEmitter &) = delete;
  SIMCCodeEmitter &operator=(const SIMCCodeEmitter &) = delete;

  /// \brief Encode the instruction and write it to the OS.
  void encodeInstruction(const MCInst &MI, raw_ostream &OS,
                         SmallVectorImpl<MCFixup> &Fixups,
                         const MCSubtargetInfo &STI) const override;

  /// \returns the encoding for an MCOperand.
  uint64_t getMachineOpValue(const MCInst &MI, const MCOperand &MO,
                             SmallVectorImpl<MCFixup> &Fixups,
                             const MCSubtargetInfo &STI) const override;

  /// \brief Use a fixup to encode the simm16 field for SOPP branch
  ///        instructions.
  unsigned getSOPPBrEncoding(const MCInst &MI, unsigned OpNo,
                             SmallVectorImpl<MCFixup> &Fixups,
                             const MCSubtargetInfo &STI) const override;
};

} // end anonymous namespace

MCCodeEmitter *llvm::createSIMCCodeEmitter(const MCInstrInfo &MCII,
                                           const MCRegisterInfo &MRI,
                                           MCContext &Ctx) {
  return new SIMCCodeEmitter(MCII, MRI, Ctx);
}

// Returns the encoding value to use if the given integer is an integer inline
// immediate value, or 0 if it is not.
template <typename IntTy>
static uint32_t getIntInlineImmEncoding(IntTy Imm) {
  if (Imm >= 0 && Imm <= 64)
    return 128 + Imm;

  if (Imm >= -16 && Imm <= -1)
    return 192 + std::abs(Imm);

  return 0;
}

static uint32_t getLit16Encoding(uint16_t Val, const MCSubtargetInfo &STI) {
  uint16_t IntImm = getIntInlineImmEncoding(static_cast<int16_t>(Val));
  if (IntImm != 0)
    return IntImm;

  if (Val == 0x3800) // 0.5
    return 240;

  if (Val == 0xB800) // -0.5
    return 241;

  if (Val == 0x3C00) // 1.0
    return 242;

  if (Val == 0xBC00) // -1.0
    return 243;

  if (Val == 0x4000) // 2.0
    return 244;

  if (Val == 0xC000) // -2.0
    return 245;

  if (Val == 0x4400) // 4.0
    return 246;

  if (Val == 0xC400) // -4.0
    return 247;

  if (Val == 0x3118 && // 1.0 / (2.0 * pi)
      STI.getFeatureBits()[AMDGPU::FeatureInv2PiInlineImm])
    return 248;

  return 255;
}

static uint32_t getLit32Encoding(uint32_t Val, const MCSubtargetInfo &STI) {
  uint32_t IntImm = getIntInlineImmEncoding(static_cast<int32_t>(Val));
  if (IntImm != 0)
    return IntImm;

  if (Val == FloatToBits(0.5f))
    return 240;

  if (Val == FloatToBits(-0.5f))
    return 241;

  if (Val == FloatToBits(1.0f))
    return 242;

  if (Val == FloatToBits(-1.0f))
    return 243;

  if (Val == FloatToBits(2.0f))
    return 244;

  if (Val == FloatToBits(-2.0f))
    return 245;

  if (Val == FloatToBits(4.0f))
    return 246;

  if (Val == FloatToBits(-4.0f))
    return 247;

  if (Val == 0x3e22f983 && // 1.0 / (2.0 * pi)
      STI.getFeatureBits()[AMDGPU::FeatureInv2PiInlineImm])
    return 248;

  return 255;
}

static uint32_t getLit64Encoding(uint64_t Val, const MCSubtargetInfo &STI) {
  uint32_t IntImm = getIntInlineImmEncoding(static_cast<int64_t>(Val));
  if (IntImm != 0)
    return IntImm;

  if (Val == DoubleToBits(0.5))
    return 240;

  if (Val == DoubleToBits(-0.5))
    return 241;

  if (Val == DoubleToBits(1.0))
    return 242;

  if (Val == DoubleToBits(-1.0))
    return 243;

  if (Val == DoubleToBits(2.0))
    return 244;

  if (Val == DoubleToBits(-2.0))
    return 245;

  if (Val == DoubleToBits(4.0))
    return 246;

  if (Val == DoubleToBits(-4.0))
    return 247;

  if (Val == 0x3fc45f306dc9c882 && // 1.0 / (2.0 * pi)
      STI.getFeatureBits()[AMDGPU::FeatureInv2PiInlineImm])
    return 248;

  return 255;
}

uint32_t SIMCCodeEmitter::getLitEncoding(const MCOperand &MO,
                                         const MCOperandInfo &OpInfo,
                                         const MCSubtargetInfo &STI) const {
  int64_t Imm;
  if (MO.isExpr()) {
    const auto *C = dyn_cast<MCConstantExpr>(MO.getExpr());
    if (!C)
      return 255;

    Imm = C->getValue();
  } else {

    assert(!MO.isFPImm());

    if (!MO.isImm())
      return ~0;

    Imm = MO.getImm();
  }

  switch (OpInfo.OperandType) {
  case AMDGPU::OPERAND_REG_IMM_INT32:
  case AMDGPU::OPERAND_REG_IMM_FP32:
  case AMDGPU::OPERAND_REG_INLINE_C_INT32:
  case AMDGPU::OPERAND_REG_INLINE_C_FP32:
    return getLit32Encoding(static_cast<uint32_t>(Imm), STI);

  case AMDGPU::OPERAND_REG_IMM_INT64:
  case AMDGPU::OPERAND_REG_IMM_FP64:
  case AMDGPU::OPERAND_REG_INLINE_C_INT64:
  case AMDGPU::OPERAND_REG_INLINE_C_FP64:
    return getLit64Encoding(static_cast<uint64_t>(Imm), STI);

  case AMDGPU::OPERAND_REG_IMM_INT16:
  case AMDGPU::OPERAND_REG_IMM_FP16:
  case AMDGPU::OPERAND_REG_INLINE_C_INT16:
  case AMDGPU::OPERAND_REG_INLINE_C_FP16:
    // FIXME Is this correct? What do inline immediates do on SI for f16 src
    // which does not have f16 support?
    return getLit16Encoding(static_cast<uint16_t>(Imm), STI);

  case AMDGPU::OPERAND_REG_INLINE_C_V2INT16:
  case AMDGPU::OPERAND_REG_INLINE_C_V2FP16: {
    uint16_t Lo16 = static_cast<uint16_t>(Imm);
    assert(Lo16 == static_cast<uint16_t>(Imm >> 16));
    uint32_t Encoding = getLit16Encoding(Lo16, STI);
    assert(Encoding != 255 && "packed constants can only be inline immediates");
    return Encoding;
  }
  default:
    llvm_unreachable("invalid operand size");
  }
}

void SIMCCodeEmitter::encodeInstruction(const MCInst &MI, raw_ostream &OS,
                                       SmallVectorImpl<MCFixup> &Fixups,
                                       const MCSubtargetInfo &STI) const {
  verifyInstructionPredicates(MI,
                              computeAvailableFeatures(STI.getFeatureBits()));

  uint64_t Encoding = getBinaryCodeForInstr(MI, Fixups, STI);
  const MCInstrDesc &Desc = MCII.get(MI.getOpcode());
  unsigned bytes = Desc.getSize();

  for (unsigned i = 0; i < bytes; i++) {
    OS.write((uint8_t) ((Encoding >> (8 * i)) & 0xff));
  }

  if (bytes > 4)
    return;

  // Check for additional literals in SRC0/1/2 (Op 1/2/3)
  for (unsigned i = 0, e = MI.getNumOperands(); i < e; ++i) {

    // Check if this operand should be encoded as [SV]Src
    if (!AMDGPU::isSISrcOperand(Desc, i))
      continue;

    // Is this operand a literal immediate?
    const MCOperand &Op = MI.getOperand(i);
    if (getLitEncoding(Op, Desc.OpInfo[i], STI) != 255)
      continue;

    // Yes! Encode it
    int64_t Imm = 0;

    if (Op.isImm())
      Imm = Op.getImm();
    else if (Op.isExpr()) {
      if (const auto *C = dyn_cast<MCConstantExpr>(Op.getExpr()))
        Imm = C->getValue();

    } else if (!Op.isExpr()) // Exprs will be replaced with a fixup value.
      llvm_unreachable("Must be immediate or expr");

    for (unsigned j = 0; j < 4; j++) {
      OS.write((uint8_t) ((Imm >> (8 * j)) & 0xff));
    }

    // Only one literal value allowed
    break;
  }
}

unsigned SIMCCodeEmitter::getSOPPBrEncoding(const MCInst &MI, unsigned OpNo,
                                            SmallVectorImpl<MCFixup> &Fixups,
                                            const MCSubtargetInfo &STI) const {
  const MCOperand &MO = MI.getOperand(OpNo);

  if (MO.isExpr()) {
    const MCExpr *Expr = MO.getExpr();
    MCFixupKind Kind = (MCFixupKind)AMDGPU::fixup_si_sopp_br;
    Fixups.push_back(MCFixup::create(0, Expr, Kind, MI.getLoc()));
    return 0;
  }

  return getMachineOpValue(MI, MO, Fixups, STI);
}

uint64_t SIMCCodeEmitter::getMachineOpValue(const MCInst &MI,
                                            const MCOperand &MO,
                                       SmallVectorImpl<MCFixup> &Fixups,
                                       const MCSubtargetInfo &STI) const {
  if (MO.isReg())
    return MRI.getEncodingValue(MO.getReg());

  if (MO.isExpr() && MO.getExpr()->getKind() != MCExpr::Constant) {
    const auto *Expr = dyn_cast<MCSymbolRefExpr>(MO.getExpr());
    MCFixupKind Kind;
    if (Expr && Expr->getSymbol().isExternal())
      Kind = FK_Data_4;
    else
      Kind = FK_PCRel_4;
    Fixups.push_back(MCFixup::create(4, MO.getExpr(), Kind, MI.getLoc()));
  }

  // Figure out the operand number, needed for isSrcOperand check
  unsigned OpNo = 0;
  for (unsigned e = MI.getNumOperands(); OpNo < e; ++OpNo) {
    if (&MO == &MI.getOperand(OpNo))
      break;
  }

  const MCInstrDesc &Desc = MCII.get(MI.getOpcode());
  if (AMDGPU::isSISrcOperand(Desc, OpNo)) {
    uint32_t Enc = getLitEncoding(MO, Desc.OpInfo[OpNo], STI);
    if (Enc != ~0U && (Enc != 255 || Desc.getSize() == 4))
      return Enc;

  } else if (MO.isImm())
    return MO.getImm();

  llvm_unreachable("Encoding of this operand type is not supported yet.");
  return 0;
}